Dimension reduction method and system for hyperspectral image

Abstract

The invention discloses a hyperspectral image dimension reduction method and a system, and the method comprises the steps: calculating k nearest neighbors (xi1, xi2,... Xik) nearest to xi according toan input hyperspectral image sample set D = {x1, x2,... Xm}, a nearest neighbor number k and a dimension d obtained through dimension reduction; calculating a local covariance matrix Zi of the hyperspectral image sample set D; calculating a weight coefficient vector Wi according to the local covariance matrix Zi; forming a weight coefficient matrix W according to the weight coefficient vector Wi,and calculating according to W to obtain a matrix M = (I-W) (I-W) T; calculating first d + 1 characteristic values of the matrix M and characteristic vectors {y1, y2,..., yn} corresponding to the d +1 characteristic values according to the matrix M; and according to the second feature vector to the (d + 1) th feature vector, forming a matrix, namely an output low-dimensional hyperspectral imagesample set matrix D '= {y2, y3,... Yd + 1}. The system comprises a nearest neighbor calculation module, a local covariance matrix calculation module, a weight coefficient vector calculation module, amatrix M calculation module and a low-order hyperspectral image sample set matrix calculation module. According to the method, dimension reduction can be carried out on the hyperspectral image throughrelatively low operation complexity, and the processing effect is ideal.

Description

technical field [0001] The invention relates to the technical field of hyperspectral image processing, in particular to a dimensionality reduction method and system for hyperspectral images. Background technique [0002] Dimensionality reduction of hyperspectral images is an important research topic in the field of hyperspectral image processing. As we all know, hyperspectral image is a kind of three-dimensional image data cube, which has rich spectral information and spatial information, and can better analyze the subtle differences between different ground objects. However, the original hyperspectral image contains hundreds of bands and a large amount of data, which contains a large amount of redundant information, which brings great difficulties and influences to subsequent work such as target anomaly detection. [0003] At present, for the dimensionality reduction processing of hyperspectral images, the conventional processing methods are band selection and feature extr...

AI Technical Summary

Problems solved by technology

[0004] Existing algorithms also achieve dimensionality reduction for hyperspectral images by finding the global optimal solution for all samples. When the amount of data is large or the sample dimension is high, the amount of calculation is very large and the calculation time is long; for the original Many correlated information in the hyperspectral image are recombined into a new set of irrelevant information to replace the original information, but the obtained dimensionality reduction results cannot reflect the hidden nonlinear properties between the sample points. It is still difficult to estimate the number of principal components

Images